Thermal Consistency Challenges in Testing High Wattage Enterprise SSDs

Transcription

1 Thermal Consistency Challenges in Testing High Wattage Enterprise SSDs Scott West, Advantest Arnaud Stehle, Mei-Mei Su, Roland Wolff, Advantest Santa Clara, CA 1

2 Thermal Consistency for Endurance Test of SSDs Endurance test definitions require thermal consistency across SSDs during entire test ±5 o C per JEDEC (JESD218A) PCIe SSD power spec is up to 25W 4X to 5X more than some SATA SSDs Test setup with single DUT thermal control is ideal, but expensive Multi-DUT thermal chambers are more cost effective, but introduce thermal consistency challenges This presentation explores these challenges Santa Clara, CA 2

3 Why Thermal Stress Test For endurance testing, thermal stress accelerates the time to fail (greatly shortening test time) Acceleration adheres to the Arrhenius equation Arrhenius Equation Predicting Temperature Dependence on Time to Fail t f = Ae E A/kT t f : time to fail A: acceleration factor E A : activation energy; T: temperature k: Boltzmann s constant This covers many failure modes of electronics but not, for example, failures causes by mechanical fatigue Santa Clara, CA 3

4 Temperature, Test Time Relationship During Endurance Testing Stress Test Time 1 Stress Temperature JEDEC Standard 218 uses the Arrhenius Equation in this form for calculations of temperature-accelerated stress times: t S FH S Ae E A/kT S.H + 1 FH S Ae E A/kT S.L Or to show the stress test time: t U FH U Ae E A/kT U.H + 1 FH U Ae E A/kT U.L t S t U FH U Ae E A/kT U.H + 1 FH U Ae E A/kT U.L FH S Ae E A/kT S.H + 1 FH S Ae E A/kT S.L From JESD218A Annex B assumes no added delays Relevant to us t S F(T S,H ) A = constant scaling factor (this drops out of the calculations) t = time (in any units as long as all t values are in the same units) T = Temperature in o K E A = Activation energy, assumed to be 1.1 ev k = Boltzmann s constant, ev/ o K FH = Fraction of time spent at high temperature S = Subscript denoting the endurance stress itself U = Subscript denoting the use condition (enterprise vs. client) H = Subscript denoting the high temperature of interest L = Subscript denoting the low temperature of interest Santa Clara, CA 4

5 Endurance Acceptance Criteria Sample size Number of DUTs tested Test Time and Temperature TBW Rating must be met, with thermal acceleration If >1000 hrs, then we can use 1000 hrs test + extrapolation (per JESD218A) Criteria (how many fails are allowed) FFR and UBER met with 60% statistical confidence UCL(functional_failures) FFR SS UCL(data_errors) min (TBW, TBR) UBER SS where FFR (Functional Failure Rate) and UBER (Uncorrectable Bit Error Rate) functional_failures is the acceptable number of functional failures data_errors is the acceptable number of data errors TBW is the endurance rating in terabytes written TBR is the number of TB read SS is the sample size in number of drives UCL() is the upper confidence limit (used to determine # fails allowed) For a zero-failure acceptance plan, UCL=0.92 From JESD218A Santa Clara, CA 5

6 Valuating Thermal Consistency In the range of typical 1000 hr RDT test: 1 o C=100 hours test time Does this mean thermal consistency improvement of ±1 o C 50 hours of test time? No, unless your customer says so! But a consistency of worse than ±5 o C will require a longer test time, lower UCL, or invalidate test results JEDEC spec is based on ±5 o C The apparatus required for this test shall consist of a controlled temperature chamber capable of maintaining the specified temperature conditions to within ±5 o C - JESD218A Section 8.1 Actual endurance stress hours Temperature ( o C) o C o C o C o C o C o C From JESD218A Table 4 If reaching TBW rating takes 1000 hrs, then 1000 hrs. plus extrapolation is still used Santa Clara, CA 6

7 Why Thermal Consistency is Important Maintain quality with manufacturing consistency Inconsistent thermal conditions create potentially unrepeatable test results for SSDs and headaches for SSD Engineers Meet conditions for Reliability Demonstration Tests JEDEC specifies ±5 o C for test equipment* * Actual conditions determined by manufacturer and customer/buyer: Alternative requirements and acceptance criteria are up to the manufacturer and purchaser to agree upon. JESD218A Violating thermal consistency requirements during a Qual might mean starting over! Longer time to market = less market share, lower margins Santa Clara, CA 7

8 Factors Affecting Thermal Consistency in Multiple DUT Chamber Chamber Performance Factors Total air flow and temperature Air guides and baffles DUT count, locations, and spacing DUT power consumption Power generated = heat; heat must be removed Worst case is with all DUTs at full power New PCIe 3.0 DUTs can be 25W compared to <10W for SATA Santa Clara, CA 8

9 Multi-DUT Chamber Considerations DUT Spacing = 1 DUT DUTs positioning perpendicular to airflow Too close: thermal disturbance between DUTs Too far apart: more expensive (floorspace) Need to balance spacing with airflow and air temperature Thermal disturban ce DUT positioning inline with air flow A gradient in temperature will occur Need to balance airflow and number of DUTs in series / temperature gradient Santa Clara, CA 9

10 Chiller Coil Baffles Multi-DUT Chamber Considerations Vertical Positioning Airflow loops through chamber to the compressor Baffles are needed to guide air into the chamber evenly Here is one scenario for 25W DUTs to meet ±5 o C 4 levels per chamber 8 DUTs deep 4 DUTs long Note: two chambers per 256 DUT system Poor baffling or too many vertical layers in a single chamber causes vertical temperature gradient Santa Clara, CA 10

11 Things We Can Control Airflow Consistency: Baffles, Empty Sockets Fluid dynamics of air flow for electronics: Pressure (V)= I Wind Resistance (R) Low resistance path Empty inline air channel = inconsistent airflow Empty perpendicular air channel = OK Low resistance path Empty space between trays = inconsistent airflow Baffles restrict air flow (e.g. half-height vs. full-height cards) Santa Clara, CA 11

12 Multi-DUT Thermal Chamber Design Air flow is complex, Sophisticated simulation, including baffles and DUT form factors is necessary to aid chamber design including baffles and DUT form factors Thermal chamber airflow simulation Santa Clara, CA 12

13 DUT Power Effects Measurement Setup Two chambers with 128 DUTs each 4 DUTs in line with air flow, 8 DUTs deep, 8 DUTs high Resistor/thermal sensor DUTs Form Factor HHHL PC Card Control power output of resistor to simulate DUT power Airflow, baffles, set to achieve ±5 o C with full load of 256 DUTs x 25W = 6.4kW Airflow between 1100 lfm and 1700 lfm Set point does not affect consistency (within operating range) Set point is the air temp, device temp is much higher 2 trays each with 2 rows of 8 DUTs Test DUT with resistor and thermal sensor Santa Clara, CA 13

14 DUT Power Sensitivity DUT Power vs. Thermal Consistency +/-1.1C +/- 1.5C +/- 3.6C +/- 5.0C Set point does not affect consistency (in operating range) Set point is air temp, device temp is much higher DUTs at 12W, 6.4kW total at ±3.7 o C ±3.6 o C ±1.1 o C Time (min) DUTs at 25W, 6.4kW total at ±5 o C ±5 o C DUTs at 10W, 6.4kW total at ±1.5 o C ±1.5 o C ±1.1 o C Time (min) Time (Min) Santa Clara, CA 14

15 Summary DUT power consumption poses thermal challenges to test High-wattage PCIe SSDs can consume 4-5x power of previous SATA SSDs Lower DUT power will have better thermal consistency Thermal consistency suffers, affecting endurance test parameters Cooling must accommodate higher power devices to prevent thermal runaway Thermal chamber design and usage is critical to meeting test criteria Come visit Advantest at booth # Santa Clara, CA 15